Sun’s New Super Computer Equals 1/20 of Human Brain; Spare Brain by 2020

By Angsuman Chakraborty, Gaea News Network
Wednesday, June 27, 2007

PleoSun Microsystems announced today that its hardware will power the largest supercomputer ever built, with 62,976 CPU cores, 125 terabytes of memory, 1.7 petabytes of disk space, and 504 teraflops of performance.

Ranger will be hosted at the Texas Advanced Computing Center at the University of Texas, Austin. It is due to go online on January 1, 2008. Ranger costs $30 million in hardware alone, and an additional $29 million for staffing and maintenance. It is funded by a grant from the National Science Foundation. link


The first supercomputer capable of crunching through a thousand trillion mathematical operations every second has been announced by IBM. This is roughly equivalent to the combined processing power of a 2.4-kilometre-high pile of laptop computers.

Blue Gene/P will be capable of a peak performance of 3000 trillion calculations, or floating point operations, per second (3 petaflops). But its sustained performance is expected to level out at around 1 petaflop.

The IBM machine’s capacity shifts the date earlier for a single, huge and very expensive, machine able to fully simulate the human brain to ~2012……software notwithstanding, of course.

Human Brain capacity
A human brain capacity is approximately 10,000 TeraFlops = 10 Petaflop = number of neurons X average number of connections between neurons X neuron firing capacity per second). Ranger will be 1/20 of that capacity.

How much is 1 petabyte exactly?
To give you an overview of the numbers involved:
1000 - kilo
1000000 - mega
1000000000 - giga
1000000000000 - tera
1000000000000000 - peta

Validity of Moore’s Law in 10 years
It is apparent that Moore’s law will hold good for at least another 10 years thanks to two promising technology and materials hafnium and graphene.

Speaking of a reinvigorated Moore’s Law, Intel’s regional vice president for sales and marketing, John Antone said in Bangkok that Intel’s latest 45-nanometer transistor had a gate that used a new element called hafnium that significantly improved the electrical characteristics by cutting down leakage current.

“So in these 45 nanometer transistors we actually deposited hafnium…It’s more like a metal, so the switching capabilities are significantly better than the capacitance associated with silicon dioxide.”

Antone said that Gordon Moore himself had called this “the biggest breakthrough in processor technology in 40 years”.

“We’ve demonstrated the 45 nanometer, which will begin production in the second half of this year, we’ve talked about and shown pictures of 35 nanometers. I believe the we’ve put a road map out for 10 years that goes down below 10 nanometers,” he said. link

There is another very exciting nano-technology material, graphene, which is strictly two dimensional crystals something like unrolled carbon nanotubes. Graphene can act as good transistors at 10 nanometers where silica currently has problems below 15 nanometers. link

When can computers equal human brain?
So, with doubling of processor capacity every 18 months in 7 or 8 years, by around 2015, individual supercomputers will have the calculating capacity of a human brain.

5 years later or so – 2020 - we will be spending $1,000 on a second brain, so we’ll have two heads, which are supposed to be better than one ;)

Pleo for now
For right now, we’ll just have to settle for the $350 Pleo from the inventor of the Furby.

Original Author: Joel B


December 20, 2007: 8:15 pm

Interesting. Does the supercomputer use cell processors? If you just used 500 ps3s, then you could achieve 500 teraflops with only about 250 thousand dollars… seems like a cheaper alternative.

jose hurtado
June 28, 2007: 8:39 pm

Interesting article. However you are making a huge mistake here, comparing pears to apples, and as a result brain processing capacity has been severely underestimated. Instead of just counting “firing” events (aka “action potentials”) in the brain as flops, as you do, you should consider how much processing it takes to compute each one of these events and multiply that by the number you got. An thats actually quite a lot. It’s humungus. In fact, your yr 2020 workstation won’t even approach a single neuron. The “transistors” of the brain are not neurons, but ion channels, receptors, g proteins, molecular motors, proteins in the synaptic complex etc.. and there are quite a lot of those within a neuron.

will not be displayed